作者:阿门路亚_ | 来源:互联网 | 2023-09-04 20:03
please make sure the input is described as a layer in caffe.prototxt
Unknown bottom blob 'data' (layer 'ConvNdBackward1', bottom index 0)
prototxt:
name: "pytorch"
input: "data"
input_dim: 1
input_dim: 3
input_dim: 256
input_dim: 256
layer {
name: "ConvNdBackward1"
type: "Convolution"
bottom: "data"
top: "ConvNdBackward1"
convolution_param {
num_output: 128
pad_h: 1
pad_w: 1
kernel_h: 4
kernel_w: 4
stride: 2
dilation: 1
}
}
...
该提问来源于开源项目:PaddlePaddle/Paddle-Lite
i found the solution, because of the caffe2mdl can not parse kernel_h & kernel_w, make kernel_h/kernel_w-> kernel_size, it works!